Feature request: Extracting causal relationships

[alinapopaqb]

Description

Would be nice if created data can be used as ground truth for Causal Discovery methods or different explainers.

In literature, there are approaches when a DAG is fitted, then used to sample data. This DAG is then considered the ground truth and a series of algorithms are benchmarked against the existing causal relationships.

Is a similar approach possible with CTGAN? If yes, would be nice to have it.

[csala]

I'm afraid this would not be possible within CTGAN, but it's something that we can definitely look at as a feature for SDV, so I'm transferring the issue there.

[anurag-ae2024]

@csala Any update on this feature?

[npatki]

Hi @anurag-ae2024, we (the SDV team), prioritize features based on demand and importance to use cases. As there hasn't been much conversation on this issue, it hasn't been on our radar.

To help us prioritize, could you describe your use case a bit more? What kind of project are you working on that will require extracting causal relationships? How do you hope to use any learned causal relationships? Thanks.